A Pragmatic Bayesian Adaptive Trial Design Based on the Value of Information: The Value-Driven Adaptive Design

Background Clinical trial designs are typically narrowly focused on error control in hypothesis testing, but this approach is inadequate in many contexts, particularly when a decision maker intends to, or must, consider multiple relevant clinical and health economic outcomes under uncertainty. Value-of-information (VoI) metrics can be used to estimate the monetary value of data collection to the decision maker. Adaptive trial designs use prespecified decision rules as data are collected and analyzed to modify the ongoing trial design. To date, VoI considerations have rarely been integrated into this approach, partly due to the computational burden. Methods We propose a value-driven adaptive design that refocuses trial design on VoI as a metric to direct trial adaptations. Specifically, a VoI analysis is performed at each interim analysis to determine whether or not the trial should proceed to the next analysis (i.e., determine whether further data collection is sufficiently valuable). We provide methods to compute the expected net benefit of perfect information, expected net benefit of sampling (ENBS) for the next analysis, and the ENBS for subsequent sequential analyses. Our approach is flexible to any statistical model, decision model, and research cost function and does not require distributional assumptions about the net benefit. Results We describe our method in detail and demonstrate its implementation via a case study comparing infant immunoprophylaxis and maternal vaccination to prevent respiratory syncytial virus–related medical attendances. Conclusions Our value-driven adaptive design aligns pragmatic clinical trial design with the requirements of decision makers. Designs with VoI-based adaptations have the potential to improve the cost-effectiveness of clinical trials. Highlights Our value-driven adaptive design is a new method that uses the expected net benefit of sampling to define stopping rules at interim analyses (i.e., to determine if further data collection is sufficiently valuable). Our method orients trial designs to efficiently produce evidence to inform the decision maker.